Dealer AI Compliance: Where the Real Risk Hides

A $412 payment quote in a text thread can turn into a six-figure problem if the customer screenshots it, walks into the box, and nobody can explain where the number came from. I have watched stores argue over worse for less money. The salesperson says the tool generated it. The BDC says the manager approved the campaign. The manager says the vendor set the rules. None of that matters much when the customer’s attorney has the message.

That is the part of AI adoption dealers need to get serious about. Not the conference-stage version with robots and magic closing rates. The real version is already running in your store: chat replies, automated service texts, trade-in offers, lead scoring, payment estimators, call summaries, email drafts, equity mining campaigns, and pricing recommendations.

CBT News recently covered the topic on Inside Automotive with Michael Affronti, and the useful thread was this: compliance and AI cannot live in separate departments. I agree. If the AI touches a customer, it is not just an IT project. It is a sales process, advertising process, F&I process, and service process all at once.

The compliance problem is not AI. It is delegation.

Dealers have always delegated work to technology. Desking tools calculate payments. CRMs send templates. Inventory tools syndicate prices. None of that is new. What is different with AI is that the tool can now create language, interpret intent, prioritize customers, and sometimes send the next message without a human reading it first.

That changes the risk profile. A bad template is usually one bad template. A poorly governed AI workflow can create 300 slightly different bad promises by Thursday afternoon.

Look, most dealers I talk to are not trying to trick anybody. They are trying to answer faster, reduce BDC load, find more cars from the drive, and keep customers from drifting to shared lead providers or auction lanes. Fair. But regulators, state AGs, plaintiff attorneys, and OEM field people do not grade intent. They look at the claim, the disclosure, the consent trail, and the customer impact.

If an AI system can send a customer-visible number — price, payment, payoff estimate, trade value, discount, rate, or appointment charge — treat that number like it belongs on a buyer’s order. Source it, timestamp it, and assign an owner.

The CARS Rule fight did not make pricing transparency optional

Some operators took the court fight over the FTC’s CARS Rule as a signal that the pressure was off. I would not run my store that way. The Fifth Circuit vacated the rule, but that did not erase federal unfair-and-deceptive authority, state advertising laws, TCPA exposure, OEM program requirements, or the simple fact that digital records are easier to subpoena than a desk manager’s memory.

The bigger issue is customer expectation. Shoppers now assume the number they see online, in chat, or in a text thread means something. If your AI assistant says “we can get you close to $399,” and the customer lands at $486 plus adds, the store owns the gap. Maybe the tool phrased it. Maybe the manager never approved that wording. The customer does not care.

I have been wrong before on how fast stores would adopt some of this stuff. I thought more dealers would slow-walk AI because of fear around compliance. Instead, I am seeing the opposite: stores are turning it on in pockets, especially in BDC and service, because the labor math is too tempting. That makes governance more important, not less.

Use the AI Exposure Ladder before you turn anything loose

Here is the framework I use when I look at a store’s AI setup. Do not start by asking whether the vendor is “compliant.” That word gets abused. Start by asking what rung of the ladder the tool is standing on.

Listening: The tool summarizes calls, reads chats, flags sentiment, or organizes notes. Lower risk, but still a data-handling issue.
Drafting: The tool writes a suggested email, SMS, or service response that a human approves before sending. Manageable if approvals are real.
Sending: The tool automatically contacts customers. Now consent, opt-out handling, timing, and message content matter a lot more.
Quoting: The tool provides or implies a payment, trade value, payoff range, discount, fee, APR, service estimate, or acquisition offer. High risk.
Deciding: The tool determines who gets contacted, what offer they receive, or how they are prioritized. This is where bias, fairness, and auditability enter the room.

Most stores are comfortable at rung one and two. The money is usually at rung three and four. That is also where the headaches live.

Service lane acquisition is a good example. If you are texting customers with equity or buyback offers, the workflow can be powerful. You already know the customer, the VIN, the service history, and often the payoff situation. That beats chasing the same auction car as every other dealer in your market. But if the SMS says “we’ll buy your Explorer for $28,000” and your used car manager meant “up to $28,000 after inspection,” you just created a trust problem and possibly a compliance problem.

The guardrails should be boring on purpose

Good AI governance in a dealership is not sexy. It is a set of permissions, review rules, and logs that keep a fast tool from outrunning the desk.

For customer communication and service lane automation, I would want five controls before scaling anything:

Approved claim library: Payment language, trade language, service estimate language, and discount wording should come from the store, not free-form AI imagination.
Human approval for numbers: Any first-pass acquisition offer, payment quote, or price adjustment should be reviewed before it hits the customer unless the rules are extremely tight.
Consent and opt-out audit: You need to prove why that customer received that message and show that STOP actually stops the workflow.
Exception routing: If the customer mentions attorney, complaint, discrimination, credit denial, accident, lemon, fraud, or “you promised,” the system should stop selling and escalate.
Transcript retention: If it is customer-facing, save it in a place your managers can actually retrieve without opening a support ticket.

Dealers using tools like AutoRelay for SMS-based service drive acquisition should be asking these questions before volume ramps up. The upside is real — more owned-channel inventory conversations, less dependence on auction buys, and better timing on customers who are already in your lane. But the message logic has to match the authority structure inside the store.

That last part is where stores get sideways. A BDC agent may be allowed to set an appointment but not quote a buy figure. A service advisor may be allowed to mention interest in the vehicle but not discuss payoff. A used car manager may approve an offer, but only after photos, history, and recon assumptions are checked. Your AI workflow needs to know those lines.

Run this audit before your next AI pilot expands

Pull 25 customer interactions from any AI-assisted channel: chat, SMS, email, service follow-up, trade acquisition, or BDC response. Score each one with a simple pass/fail on seven items:

Was there a clear consent basis for the contact?
Did the message include a price, payment, rate, trade value, fee, or service estimate?
If a number appeared, can you identify the source?
Did the message overpromise availability, discount, financing, or timing?
Was opt-out language present where it needed to be?
Could a manager retrieve the full thread in under two minutes?
Was there a clear human owner for the interaction?

If more than three of the 25 fail, do not expand the tool yet. Tighten the rules first. AI can absolutely help dealers communicate faster and source better inventory from their own customers. But speed without control just creates cleaner evidence of sloppy process.

See how AutoRelay helps dealers acquire inventory from their own service drive → getautorelay.com